Mostafai, Fatemeh

A Systematic Review of AI-Based Speech Assessment (2010-2025): Assessment Criteria an‎d Research Contexts in Focus

Applying artificial intelligence-based technologies in analyzing an‎d eva‎luating human speech for different applications like language learning, detection of speech disorders, an‎d promoting human-AI interaction has been developing during the last decades. To synthesize the available literature, this study intended to review the existing literature on the automatic speech assessment considering the role of different contributing speech assessment criteria, accumulating previously stated contexts for AI-based speech assessment based on existing English-published investigations of related fields between 2010-2025. Following PRISMA 2020 guideline, among 163 initial records, 113 records were excluded due to the reasons mentioned in the study an‎d based on the Inclusion an‎d exclusion criteria. Finally, 50 studies were identified an‎d analyzed. According to the findings of this review, among three major speech assessment features like linguistic, paralinguistic, an‎d pragmatic criteria, the role of paralinguistic features was dominant due to their measurable nature. Linguistic criteria played significant role that reflects recent developments in speech recognition an‎d automatic speech eva‎luation systems. Nevertheless, the limited presence of pragmatic features across the reviewed studies indicated the complication of using pragmatic aspects of speech in computational situations. Pragmatic features contain context-sensitive interpretation, speaker intention, an‎d appropriateness within communicative background, which are less directly visible in the speech signal that typically need human subjective judgments. Furthermore, educational, clinical, an‎d AI-development contexts were the most frequent mentioned settings for included studies. The majority of reviewed studies were conducted in educational contexts with assessment tasks an‎d data collection procedures aligned with the proposed application domains. The findings of current study can lead future researchers to develop the research domains by applying the provided classified information in different interdisciplinary fields.